Key Takeaways
- Stop treating a new date as an update. Changing when a page says it was published, with no change to what it says, gives an assistant nothing new to read. It is the most common freshness ritual and the emptiest one.
- Ask which facts actually rot. Prices, locations, staff, services, and credentials go stale. Your method, your reasoning, and your explanations mostly do not. Recency matters per fact, not per page.
- Expect a delay between your edit and the answer. A change has to be read again before it can be repeated, and every source refreshes on its own schedule with no published timetable. Weeks is normal.
- Fix the wrong fact before you write the new page. A stale claim being repeated about you costs more than an absent one. Correct what is wrong, then go back to publishing.
- Publish on a rhythm you can hold for a year. Steady beats a burst. A site that changes a little every month keeps giving crawlers a reason to come back, which a site rewritten once a year does not.
Everyone has been told that AI likes fresh content, so a lot of coaching sites now carry a date from last month and a body from three years ago. Nothing about the site has changed except the date.
That gap is worth looking at directly, because it is where most freshness advice quietly stops being true. This is a field-notes post: the patterns below come from our own repeat runs and are described in illustrative terms rather than measured client data.
The short version is that freshness is not a property of a page. It is a property of individual facts, and only some of them decay.
How fresh does content need to be for AI to use it?
Fresh enough that nothing on it has become false. Age by itself is a weak signal, and old pages that are still accurate get quoted without any apparent penalty.
What we keep seeing in repeat runs is that the pages getting quoted are rarely the newest ones. They are the ones that state something plainly and are still correct. A well-written explainer from two years ago outlives a thin post from last week, every time we look.
Where recency does bite is on facts with expiry dates. A price that changed, an address that moved, a service you stopped offering. Those produce answers that are wrong about you, which is a different and more expensive problem than being absent.
Which facts actually go stale
Sort your page by shelf life rather than by publication date. Some facts rot in months, some hold for years, and treating them the same is why freshness work feels endless and changes nothing.
Short shelf life: prices, packages, availability, location, staff, hours, current credentials, and anything with a year in it. Long shelf life: how you work, who you help, what problem you solve, why your method is what it is.
Most sites have this backwards. The explainer gets rewritten because it feels stale, and the pricing page keeps a number that stopped being true last spring. One of those changes an answer and the other does not.
What counts as a change and what does not
A change counts when the stored text of your page is different afterward and that difference carries new information. Editing for style clears the first bar and not the second.
This is the distinction that makes a content calendar honest. Four entries can all read as "updated the services page" and only two of them gave anyone anything to work with.
All four go in a content calendar as an update to the same page. The first is a date change with identical text beneath it. The second reads as real work and states nothing that was not already stated. Only the bottom two put something new in front of anyone. The pattern is illustrative, drawn from our own repeat runs, not measured client data.
| Fact | Shelf life | What a stale version causes | How often to check |
|---|---|---|---|
| Price and packages | Months | An assistant quotes a number you no longer charge | Whenever it changes, same day |
| Location and service area | Years, then suddenly wrong | You get named for the wrong city, or not at all | At every move, then yearly |
| Services offered | A year or two | Buyers arrive asking for something you dropped | Twice a year |
| Credentials and certifications | Tied to renewal dates | A credibility claim that no longer checks out | At each renewal |
| Method and philosophy | Long | Very little, if it is still what you do | Only when it genuinely changes |
Why your edit takes weeks to show up
Your change has to be fetched again before it can be repeated, and nothing about that timing is published. The lag is normal, it is not a sign the edit failed, and it is longer than almost everybody expects.
Two things drive it. Your own page has to be re-read, and so do the third-party pages that mention you, which refresh on schedules you do not control. OpenAI's crawler documentation notes that even a robots.txt change can take around a day to take effect on its side, which is a useful floor for how to think about everything slower than that. The details are in its documentation for site owners.
Some of what a model holds comes from large periodic snapshots of the web rather than a live fetch, and those are assembled on their own cycle. Common Crawl publishes its archives on a rolling schedule, which is one concrete reason a correction can be live on your site and absent from an answer for a while. Google's guidance on AI features in Search makes the same general point differently: these systems change without notice, so a fixed expectation about timing is the wrong shape of belief.
The practical consequence is to date your own work. Note what you changed and when, then re-ask the question in a month. Without that record you cannot tell a slow success from a failure. More on the underlying mechanics is in how often ChatGPT updates what it knows.
The cadence that works
Small and regular beats large and rare. A page or two of real change every month gives crawlers a standing reason to come back, and it is a rhythm one person can actually hold.
Order the work by damage. First, correct anything that is now false. Second, state something never stated. Third, improve what already works. Most people run that list backwards and wonder why nothing moves.
When an assistant is repeating something about you that stopped being true, the fix usually is not on your site at all. It is on the third-party page the claim came from, and the procedure for that is in fixing an out-of-date picture of your business in AI. The broader patterns we keep seeing across repeat runs are in what 25-question test runs keep showing us.
One honest caveat. These are patterns from our own runs in coaching and consulting niches, described in illustrative terms. They are a place to start looking rather than a rule to plan against, and your own market is the only thing that settles it for you.
If you would rather see where you stand before deciding what to rewrite, the free check runs 25 real buyer questions from your niche through ChatGPT at $0 and emails a plain-English readout within the hour. It runs as soon as you ask for it.
Questions we hear the most
How fresh does content need to be for AI to use it?
Fresh enough that the facts on it are still true. Age alone matters far less than whether anything on the page has gone wrong, and a five-year-old explainer that is still accurate gets used without difficulty.
Does changing the published date help?
On its own, no. If the text is identical, the stored copy of your page is identical, and an assistant has nothing new to say about you. It also costs you the ability to tell what actually changed.
How often should I update my website for AI?
Change something real every month or two, and correct wrong facts the day you notice them. A steady rhythm you can hold beats a large rewrite that happens once and then never again.
How long before an assistant reflects my change?
Usually weeks rather than days, and sometimes longer. Your page has to be fetched again, and the third-party pages that mention you refresh on their own schedules, which nobody publishes.
Does deleting old content help my visibility?
Rarely. Deleting removes a page an assistant could have quoted. Correcting it is almost always better, and the exception is a page describing a service you genuinely no longer offer.
What does AnswerHalo actually watch for freshness?
Whether the facts in an answer about you match the facts you publish today. When an assistant repeats something you changed a year ago, that gap is the finding, and the fix is usually a source you do not control.
Are these numbers from client accounts?
No. Everything here is a pattern from our own repeat runs, described in illustrative terms. We publish no client figures, and nothing in this post should be planned against as a measurement.
Can I run the free check today?
Yes. It runs as soon as you ask for it, and the readout lands within the hour between 7am and 5pm Central. Reading your own top three pages for facts that have quietly gone wrong is free and can happen this afternoon.
Your buyers are already asking. Find out what AI tells them.
The free check asks ChatGPT 25 real buyer questions about your niche and sends you the report within the hour, at $0. If it shows you are already getting named everywhere, we will say so in plain words.
Get my free check →